The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the vnapp-cli tool from the author's GitHub repository (github.com/cawcut/skill-vn) to the user's local directory during the installation process.
[REMOTE_CODE_EXECUTION]: The skill setup logic downloads an executable binary and runs it locally. To mitigate supply chain risks, the skill performs a SHA-256 checksum validation against a hardcoded hash (857417f671fd6f38ec30cf993f80c79efede7641f3532c2e9a99e75177a5853f) before extraction and execution.
[COMMAND_EXECUTION]: The skill uses various shell commands for environment discovery (sw_vers, uname), binary installation (curl, unzip, chmod), and tool execution. It also manages macOS Gatekeeper attributes using xattr to allow the locally downloaded CLI tool to run.
[PROMPT_INJECTION]: The skill handles untrusted data through user-provided media file paths, representing an indirect prompt injection surface.
Ingestion points: Local file paths for videos, images, and SRT subtitles provided by the user in SKILL.md instructions.
Boundary markers: The instructions mandate the use of absolute paths and require the agent to wrap these paths in double quotes in shell commands to prevent command injection.
Capability inventory: The agent can execute system commands (test, open) and the vnapp-cli binary with arguments derived from user input.
Sanitization: The skill relies on the agent's ability to properly escape and quote file paths provided by the user; no explicit input validation or regex filtering is defined in the instructions.

vn-skill